PHENOMIC PREDICTION OUTPERFORMS GENOMIC PREDICTION FOR SOME BREAD WHEAT TRAITS ACROSS DIVERSE ENVIRONMENTAL TRIALS
Author(s)
Date Issued
2025
Type
conferenceObject
Abstract
Genomic prediction (GP) is a promising tool to improve wheat breeding efficiency for complex
traits like yield and protein content, which are highly influenced by the environmental conditions.
However, despite the decreasing cost of genotyping technologies, implementing large-scale
genotyping still poses a significant financial challenge. As a cost-effective alternative, phenomic
prediction (PP) using near-infrared spectroscopy (NIRS) on whole-grain flour, already common in
breeding programs, has gained interest for predicting multiple traits.
In this study, conducted within the Agritech National Research Center, funded by MIUR
European Union PNRR – MISSIONE 4 COMPONENTE 2, INVESTIMENTO 1.4 -SPOKE 1
INVESTIMENTO 1.4, a panel of 170 recombinant inbred lines (RILs) of bread wheat (Triticum
aestivum
L.) was genotyped using 15K SNP array and evaluated for grain yield (GY), grain protein content
(GPC), grain protein deviation (GPD), grain yield deviation (GYD), plant height (PH), and heading
date (HD). Field trials were conducted under rainfed and irrigated conditions across three
growing seasons at CREA Foggia, Italy. NIRS data were collected after each growing season and
under each water regime. Adjusted phenotypic means for each trial were integrated with
genotypic and NIR-based phenomic information to develop three prediction models (PP, GP, and
GP+PP), which were tested across three different scenarios: within-environment, across
environments, and multi-environment. In addition, an optimized subset of 46 genetically diverse
lines was used as a training population to implement GP and PP models. Prediction accuracy was
assessed using two cross-validation schemes.
In the within-environment scenario, PP matched or outperformed GP for all traits, showing the
highest prediction accuracy for GPC and GPD (0.92, 0.87) compared to GP (0.24, 0.23). For other
traits, the combined GP+PP model delivered the best performance. In the across-environment
and multi-environment scenarios, PP showed superior accuracy for GPC, GPD, and PH, while GP
remained better for GY and GYD. Notably, using the fixed 46-line training population yielded
prediction accuracy comparable to k-fold cross-validation, highlighting its potential to reduce
breeding cost. Overall, PP emerges as a viable, scalable, and cost-efficient strategy to enhance
selection in wheat breeding, either alone or in combination GP.
Conference(s)
LXVIII SIGA Annual Congress